You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Relying on PR and commit titles to determine which attribute paths to build alone is sensitive to human error, and committers are depending on state of mind prone to merge PRs where CI is all green without checking if any builds were dispatched or not.
Therefore i propose some extra logic to determine which packages to build in ofborg CI:
Build packages whose meta.position is touched.
Skip this logic if more than say 5 packages share the same meta.position.
Build xyz if any file in pkgs/by-name/xy/xyz/ is touched
This is safe, since the by-name CI check ensures there are no ../* references in by-name packages.
This will cause a build if only say patch files are touched, and not only trigger for package.nix
This can of course be a fallback used only if no valid/existing attribute path could be extracted from the PR/commit titles.
Thoughts?
The text was updated successfully, but these errors were encountered:
pbsds
changed the title
Detect what to rebuild builds based touched file paths
Detect what to rebuild builds based on touched file paths
Dec 20, 2023
pbsds
changed the title
Detect what to rebuild builds based on touched file paths
Detect what to build based on touched file paths
Dec 20, 2023
Relying on PR and commit titles to determine which attribute paths to build alone is sensitive to human error, and committers are depending on state of mind prone to merge PRs where CI is all green without checking if any builds were dispatched or not.
Therefore i propose some extra logic to determine which packages to build in ofborg CI:
meta.position
is touched.meta.position
.xyz
if any file inpkgs/by-name/xy/xyz/
is touchedby-name
CI check ensures there are no../*
references inby-name
packages.package.nix
This can of course be a fallback used only if no valid/existing attribute path could be extracted from the PR/commit titles.
Thoughts?
The text was updated successfully, but these errors were encountered: